The AI Trust Ops Trustworthiness Diagnosis List presents items to be checked to secure trustworthiness according to the implementation stage of AI products and services to solve the inherent risks and technical limitations of AI and to prevent side effects in the process of utilizing and spreading. Therefore, it is possible to perform activities to secure trustworthiness according to the service fields and environments provided by a company by selecting, applying, and inspecting appropriate requirements and verification items in consideration of the company's internal technology capabilities and product service characteristics. In addition, legal and institutional aspects such as ethics and personal information protection laws are also required in addition to technical aspects in order to secure AI trustworthiness. It is required to additionally review the 「Self-Checklist for Practice of AI Ethics Standards」 to check ethical considerations in AI and the 「Artificial Intelligence (AI) Personal Information Self-Checklist」 to check compliance with personal information protection.
Procedures for using self-diagnosis of AI trustworthiness are as follows.
- Analysis of AI service risk impacts: The first thing to consider for risk impact analysis is the potential impact according to the purpose, scope, and target of the use of AI services. Even for services with a similar purpose, the degree of impact of risk may vary depending on the final intervention of a human in the reasoning result of AI.
- Selection of requirements and verification items: Based on the analysis of ‘①’, select the requirements and verification items required to secure trustworthiness in AI services. If AI services are used for public purposes, or if it is considered likely to cause irreversible damage to human bodies and properties, such as diagnosis of a disease or stock trading, it is recommended to select all possible requirements and verification items. Requirements and verification items that are deemed unnecessary can be excluded from inspection by marking ‘N/A’.
- Performance of self-inspection: Check related outputs such as procedure documents, codes, and analysis data required for determining whether each verification item is met by referring to the examples of technologies and techniques introduced in Requirement Pool and Element Technology. And perform necessary activities for items that need test or measurements.